城市森林调查是城市森林研究和规划建设的前提和基础工作,而遥感和GIS技术已成为现代城市森林调查的主要方法。遥感信息源和遥感分类技术是决定城市森林遥感调查质量的两个关键因素,研究认为,根据城市森林特点及城市森林研究和规划建设的要求,城市森林遥感调查采用的遥感影像分辨率以8¹0m为宜,神经网络模型分类技术应用于城市森林遥感调查效果较好。本文以南京市主城区为例,以10m分辨率多光谱spot卫星图像与2.5m分辨率全色spot卫星图像的融合图像作为解译信息源,采用dARTMAP神经网络模型进行监督分类,提取了一套具有较高精度的南京市主城区城市森林空间属性数据和图件,并在此基础上分析了南京市城市森林基本特点及规划建设方向。

Urban forest generally refers to the trees and related vegetations in the urban area and its surroundings.Academically established concept of urban forest marks the great change of man's opinion on forest.The urban forest concept was introduced into China in the late（1980s.） Theoretical research on and case study of urban forest becomes more and more popular. Urban forest is a rising research area.Cultivation of urban forest has been an important part in urban planning and sustainable development.Continuous development of remote sensing and GIS facilitates more rapidly and accurately urban forest investigation.Information sources and methods for remote sensing classification are two crucial factors in terms of the quality of urban forest remote sensing investigation. Researches reveal that urban forest is highly fragmentized with a number of patches of different sizes and complex boundaries,hence remote sensing images of relatively high resolution of 8¹0 m are needed.It is difficult to make urban forest remote sensing classification by traditional visual interpretation.Remote sensing classification based on neural network model can well solve such problems as uncertainty of complex boundaries and mixed pixel resolution,much suitable to urban forest remote sensing investigation. Taking the recent high-resolution remote sensing images of Nanjing city as an information source,we have made remote sensing survey of forest landscape in the main part of Nanjing city by use of dARTMAP classification method.Image interpretation is made under Matlab environment and supervised classification is made by use of dARTMAP neural network model.The accuracy analysis function of ERDAS software is employed to make accuracy evaluation of classified data.According to the accurate test confusion matrix,the total classification accuracy reaches 89.0% and kappa coefficient is up to 0.7271.And the forest landscape distribution map of the main city of Nanjing is made and the spatial attribute data of the patches deduced under Arcmap environment.